• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

DLSS Momentum Continues: 50 Released and Upcoming DLSS 3 Games, Over 250 DLSS Games and Creative Apps Available Now

I said the other day that AMD has literally never provided a meaningful upgrade to FSR and HUB have just said the exact same thing.

Slightly annoying they didn't update the dll files for DLSS3 in some of those games though as some were as old as 2.5.1 and DLSS quality control has improved a lot in later versions. It's a super easy drag and drop and something most people would easily do for the improved image quality and the comparison would be even better.

They should have mentioned this in the video as well that DLSS dll swapping is a thing whereas AMD don't support this and fsr updates require each game Dev to provide an update.
 
Last edited:
I said the other day that AMD has literally never provided a meaningful upgrade to FSR and HUB have just said the exact same thing.

Slightly annoying they didn't update the dll files for DLSS3 in some of those games though as soke were as old as 2.5.1 and DLSS quality control has improved a lot in later versions. It's a super easy drag and drop and something most people would easily do for the improved image quality and the comparison would be even better.

They should have mentioned this in the video as well that DLSS dll swapping is a thing whereas AMD don't support this and fsr updates require each game Dev to provide an update.

Not even a case of drag and drop, install dlss swapper and literally click game profile and swap :cry: But apparently, it's such a hardship and pc gamers don't tweak settings :cry:

Also, I'm pretty sure they don't turn of things like motion blur, CA, DOF, lens flare and so on which can and often does negatively impact the quality of upscaling tech.

AMD really don't have any choice now but to improve it, sony obv got fed up with lack of support and the quality of FSR hence why they have gone of to do their own one and even though fsr works on a variety of hardware, as shown by polls and even their own userbase, no one wants to resort to using it.
 

Pretty factual but brutal conclusion by Tim.
I'm so bored of these useless DLSS vs FSR comparison videos. The situation hasn't changed in years, and nor will it until AMD fundementally change the way FSR is developed. Which will no doubt be something they announce, so why do we need these constant check-ins other than the purpose of feel-good clickbait for Nvidia owners?

More importantly, where's XeSS? Why does nobody ever include it? It's THE most interesting upscaling solution out there, with the benefits of looking much better than FSR whilst also having much wider hardware compatibility than DLSS. I know from my own testing on a 1080 Ti that it's the best upscaling solution available for people without an RTX card. And yet we get videos like this complete waste of time instead of coverage of that.
 
The issue is all these reviewers are completely clueless and just yap stupidly & incessantly. Ofc FSR hasn't seen improvements... because there's no improvements to be had. It is the best vendor-agnostic TAAU solution available already, and where it is aced it is done so through a LOT of speed-ups from hw acceleration for the clamping step, which is the only real fundamental difference between all the vendor TAAUs (hence why you see ghosting/shimmering much improved in DLSS/XeSS XMX). So the only solutions available are either you spend a lot more ms to improve that step (you see this with TSR, but also in XeSS's DP4A abysmal performance and with no real notable improvements, just different IQ trade-offs) at which point you might as well just run TAA because it will be better, at least at lower res where the upscaling trade-off isn't worth it; or you have to introduce hw/acc - which is why you also see XeSS XMX is close to identical to DLSS, because there's no magic model for DLSS that makes it the bees knees, it's simply a function of the HW doing that step really fast. So then why isn't AMD doing it? Simple, because these hardware changes take years to plan and execute and they're unfortunately still reliant on making major arch changes around consoles.

Tbh instead of crying about AMD these ppl would do better to pester the devs which implement all these solutions very poorly, failing to even adhere to the implementation guide's simple steps (f.ex. lod bias mismatch, masks etc.) There's already plenty of IQ improvements that could be had but are missed because they won't do the work required.
 
Last edited:
It’s on nvidia to figure out how to ensure the latest version of DLSS is being deployed widely. No need to make excuses for them
What does this even mean? Nvidia provide the dll files to devs for new games coming out when under development, the dev then ships that version they used in the release of the game.

Maybe a dev patches the game and adds the latest dll for DLSS or maybe they don't. This has nothing to do with Nvidia but everything to do with the developer including it in later game patching if they want to

Or you know, just swap the dll manually which takes seconds.
 
Last edited:
The issue is all these reviewers are completely clueless and just yap stupidly & incessantly. Ofc FSR hasn't seen improvements... because there's no improvements to be had. It is the best vendor-agnostic TAAU solution available already, and where it is aced it is done so through a LOT of speed-ups from hw acceleration for the clamping step, which is the only real fundamental difference between all the vendor TAAUs (hence why you see ghosting/shimmering much improved in DLSS/XeSS XMX). So the only solutions available are either you spend a lot more ms to improve that step (you see this with TSR, but also in XeSS's DP4A abysmal performance and with no real notable improvements, just different IQ trade-offs) at which point you might as well just run TAA because it will be better, at least at lower res where the upscaling trade-off isn't worth it; or you have to introduce hw/acc - which is why you also see XeSS XMX is close to identical to DLSS, because there's no magic model for DLSS that makes it the bees knees, it's simply a function of the HW doing that step really fast. So then why isn't AMD doing it? Simple, because these hardware changes take years to plan and execute and they're unfortunately still reliant on making major arch changes around consoles.

Tbh instead of crying about AMD these ppl would do better to pester the devs which implement all these solutions very poorly, failing to even adhere to the implementation guide's simple steps (f.ex. lod bias mismatch, masks etc.) There's already plenty of IQ improvements that could be had but are missed because they won't do the work required.
XeSS is better than FSR in basically every game that features both upscalers as an option. XeSS works on any card too. Just the only issue is XeSS needs to be supported by a game first.
 
The issue is all these reviewers are completely clueless and just yap stupidly & incessantly. Ofc FSR hasn't seen improvements... because there's no improvements to be had. It is the best vendor-agnostic TAAU solution available already, and where it is aced it is done so through a LOT of speed-ups from hw acceleration for the clamping step, which is the only real fundamental difference between all the vendor TAAUs (hence why you see ghosting/shimmering much improved in DLSS/XeSS XMX). So the only solutions available are either you spend a lot more ms to improve that step (you see this with TSR, but also in XeSS's DP4A abysmal performance and with no real notable improvements, just different IQ trade-offs) at which point you might as well just run TAA because it will be better, at least at lower res where the upscaling trade-off isn't worth it; or you have to introduce hw/acc - which is why you also see XeSS XMX is close to identical to DLSS, because there's no magic model for DLSS that makes it the bees knees, it's simply a function of the HW doing that step really fast. So then why isn't AMD doing it? Simple, because these hardware changes take years to plan and execute and they're unfortunately still reliant on making major arch changes around consoles.

Tbh instead of crying about AMD these ppl would do better to pester the devs which implement all these solutions very poorly, failing to even adhere to the implementation guide's simple steps (f.ex. lod bias mismatch, masks etc.) There's already plenty of IQ improvements that could be had but are missed because they won't do the work required.
The stuff about XeSS really isn't accurate. It looks much better than FSR even running on an AMD/Nvidia card. I'm yet to see any evidence that there are image quality compromises in running it on an AMD/Nvidia card as opposed to Arc. The compromise comes in terms of getting less of a performance boost due to it needing to take up shader resources, but in my experience testing it on non-Arc cards, XeSS in Balanced mode looks significantly better than Quality FSR whilst performing close to it. As a result, it's the best upscaling solution available for non-RTX cards and very close to DLSS these days. I'd take it over FSR every time as an AMD/GTX user. Which also rather dismisses the argument that there's some special hardware component needed to make upscaling look good. People keep parroting the Nvidia marketing spiel that DLSS is only possible on RTX cards due to the awesome power of the tensor cores, and yet DLSS 1.9 (the first "modern" version of DLSS and a prototype of what we have now) ran on the shaders - just as XeSS does on AMD/Nvidia cards now. The hardware component on the user's end simply accelerates the performance of what is inherently a software solution. That isn't to say there aren't benefits to having dedicated hardware to offload it to (you get both the image quality AND the biggest performance uplift), but it's just not true to say you can't have DLSS-like image quality without dedicated hardware. You can. I've seen it with my own eyes running XeSS on a non-Arc card. The key to it all is the work that's done on the Intel/Nvidia side with deep learning. There's no reason AMD can't do that as well and provide an XeSS-like solution, which would look much better than FSR does now.

Of course, the disadvantage of that approach will be ditching all those people on old and low-end systems that FSR helps now. XeSS is utterly useless on cards that don't support DP4a, where it has to use a secondary fallback mode and actually performs worse than native rendering. GPUs without DP4a support include RDNA 1, Vega and Polaris, which I'd imagine covers the vast majority of current AMD GPU users. FSR as it exists today provides a nice boost for all those people stuck on Vega iGPUs or RX 580s or 5000-series cards, as well as Nvidia users on Maxwell or older architectures (though I'd say that's a less important demographic at this point). A new, XeSS-style FSR would almost certainly abandon those people. Which is why I'd like to see it stick around in its current form, even if AMD do develop something new to service people running RDNA 2/3 cards. After all, those people need the performance more than anybody.
 

Pretty factual but brutal conclusion by Tim.
It's a shame they didn't test NMS. IIRC the switch version with FSR was very well received and AFAIK they ported the work they'd done for switch to other platforms. It does make me wonder that if Hello Games managed to implement FSR as well as they have are the issues we're having with FSR just technical or also down to the devs too. Not that that makes a difference from a players perspective - a crap image is still crap regardless why.

But it'd be interesting to know if FSRs weakness lies in the technology being less capable, or if it's difficult to work with to make it perform well. And whether DLSSs strengths lie in the technology being just better, or it being more forgiving to work with. I suspect the story is somewhere in the middle with both tbh - it normally is.

Edit: here's the DF video where it shows off some FSR on NMS.
 
Last edited:
  • Like
Reactions: mrk
XeSS is better than FSR in basically every game that features both upscalers as an option. XeSS works on any card too. Just the only issue is XeSS needs to be supported by a game first.
Incorrect.
The stuff about XeSS really isn't accurate. It looks much better than FSR even running on an AMD/Nvidia card.
It doesn't. It has different trade-offs in terms of ghosting vs shimmering (as well as different artifacts/visual bugs - f.ex. been testing Cyberpunk again lately and spent considerable time with XeSS and saw visual glitches that simply aren't there with FSR, such as rain puddles being a complete mess in motion like a weird soup, the vegetation in Northside shimmering horrendously during daytime, the LODs during outdoor cutscenes going haywire, which was the case ever since they introduced XeSS etc.), and on top of that it has much higher performance penalty.
I'm yet to see any evidence that there are image quality compromises in running it on an AMD/Nvidia card as opposed to Arc.
The DP4A looks the same on all. That's now what I'm talking about.
The compromise comes in terms of getting less of a performance boost due to it needing to take up shader resources, but in my experience testing it on non-Arc cards, XeSS in Balanced mode looks significantly better than Quality FSR whilst performing close to it.
The compromise comes in the form of what AA weakness you focus on, as I've said above, XeSS allows more ghosting but will see less shimmer/edge aliasing compared to FSR 2, but the later will have a more stable "clamping" of the image overall. More importantly this is not really a difference between the two but rather a choice (for the devs) - you can setup FSR2 like XeSS as well (or vice versa).
As a result, it's the best upscaling solution available for non-RTX cards and very close to DLSS these days.
But it's not, because again - different trade-offs and on top of it, the coup de grâce is that it has much heavier performance costs, and that's what kills it. For low-end cards - it's DOA. For mid-range, it becomes usable but it's really only a question for AMD users. For high-end, only NV has such cards (so you use DLSS because it's the best). So who does it actually benefit? And that's assuming you'd want the ghosting and visual bugs of XeSS over FSR's ones.
I'd take it over FSR every time as an AMD/GTX user.
Feel free. A lot of the IQ differences come down to subjectivity anyway (tho ofc not the performance).

I just wish AMD's marketing department wasn't completely trash (since always). They could dispel a lot of these myths by making some proper video comparisons. I'd do it myself but screw that, I'd rather use that time differently and they can afford it anyway. Or they can just keep losing brand value, but hey, their problem. I just wish XeSS was as good as a lot of people hype it up to be but after testing for tens of hours in Cyberpunk it unfortunately has disappointed. For the most part I'd characterise it as: FSR has minor flaws all the time, but XeSS has some major flaws sometimes. I find consistency easier to stomach.
 
Incorrect.
Evidence? I/we have all played the games using all 3 tech, and FSR always falls behind, always. Even in games that are AMD sponsored, FSR is behind.

Here's a scene I just cooked up with all 3, Max settings path traced. The time cycle was frozen using CET and all NPC/world activity was frozen too so apart from animated billboard ads that seem to defy TIME AND SPACE!!!!!!!!!!!!!, this is as fair a comparison as you can get:

xqPQ2WC.jpeg



I chose Cyberpunk because it's the most obvious one to use since it uses all 3 vendor upscalers to as good effect as the tech is capable of.

Notes:
- From this comparison it is clear that XeSS Ultra Quality is just as good as DLSS Quality but at the cost of 15fps
- XeSS Quality vs DLSS Quality shows XeSS is 4fps behind DLSS, and is ever so slightly less sharp vs DLSS Quality, though you have to zoom into the pixels about +400% to see this
- FSR Quality vs XeSS and DLSS Quality is no contest, FSR is visibly softer even without having to zoom in.
- DLSS Quality with Ray Reconstruction (with or without Frame Gen) offers unmatched image quality and performance. Not even DLAA or native resolution can match it, especially in path tracing or ray tracing since Ray Reconstruction is only possible when the Tensor cores are active, so DLSS has to be in use. It is basically a landslide victory for the amalgamation of all of the DLSS 3.5 umbrella of technologies working together. You would get the same image using DLSS Performance at 4K as well.
- What you cannot see from the screenshots is the temporal stability of darker areas or surfaces under lighting/reflection. The pavement at the back there shimmers with RT denoiser artefacts on all but the Ray Reconstruction output, for example, where it is completely clean.
 
Last edited:
@mrk You would have done better to read again for comprehension. I explicitly talked about issues in motion. Who even tests static scenes with TAAUs? Completely pointless.

How FSR is superior to XeSS to how it handles vegetation [for this one wait for 4K to process, can't see it otherwise - or better yet, go test it yourself and verify]:

Or how some issues that exist even in TAA are then exaggerated with XeSS but get suppressed better with FSR (and this is corollary to the visual artifacts that then pop up in certain quest scenes with only XeSS as a result; repeatable in the Q for Meeting Oda, or 6th Street with Panam etc.):

I wish I had a rain scene too, because that's the worst of the bunch, and that's where XeSS is just dog doodoo. Alas, you'll have to make do with these.
{edit- Similar issue but with reflection: https://youtu.be/TV-EjAJjPhI}

I can promise you, there is no one that's spent more time testing Cyberpunk than me, and I've actually played through the whole game with these things, all quests & gigs (and every game version since 1.0, so I know how they all act and how they've changed). Actually playing, not photo mode-ing or superficial testing for the random youtube "tech" channel. That's why I'm the only guy you know who's pointed these things out. And if you had done so too then you wouldn't argue with me about what are obvious differences between these TAAUs.
 
Last edited:
@mrk You would have done better to read again for comprehension. I explicitly talked about issues in motion. Who even tests static scenes with TAAUs? Completely pointless.

How FSR is superior to XeSS to how it handles vegetation [for this one wait for 4K to process, can't see it otherwise - or better yet, go test it yourself and verify]:

Or how some issues that exist even in TAA are then exaggerated with XeSS but get suppressed better with FSR (and this is corollary to the visual artifacts that then pop up in certain quest scenes with only XeSS as a result; repeatable in the Q for Meeting Oda, or 6th Street with Panam etc.):

I wish I had a rain scene too, because that's the worst of the bunch, and that's where XeSS is just dog doodoo. Alas, you'll have to make do with these.
{edit- Similar issue but with reflection: https://youtu.be/TV-EjAJjPhI}

I can promise you, there is no one that's spent more time testing Cyberpunk than me, and I've actually played through the whole game with these things, all quests & gigs (and every game version since 1.0, so I know how they all act and how they've changed). Actually playing, not photo mode-ing or superficial testing for the random youtube "tech" channel. That's why I'm the only guy you know who's pointed these things out. And if you had done so too then you wouldn't argue with me about what are obvious differences between these TAAUs.


6edeMgK.jpeg
 
@mrk You would have done better to read again for comprehension. I explicitly talked about issues in motion. Who even tests static scenes with TAAUs? Completely pointless.

How FSR is superior to XeSS to how it handles vegetation [for this one wait for 4K to process, can't see it otherwise - or better yet, go test it yourself and verify]:

Or how some issues that exist even in TAA are then exaggerated with XeSS but get suppressed better with FSR (and this is corollary to the visual artifacts that then pop up in certain quest scenes with only XeSS as a result; repeatable in the Q for Meeting Oda, or 6th Street with Panam etc.):

I wish I had a rain scene too, because that's the worst of the bunch, and that's where XeSS is just dog doodoo. Alas, you'll have to make do with these.
{edit- Similar issue but with reflection: https://youtu.be/TV-EjAJjPhI}

I can promise you, there is no one that's spent more time testing Cyberpunk than me, and I've actually played through the whole game with these things, all quests & gigs (and every game version since 1.0, so I know how they all act and how they've changed). Actually playing, not photo mode-ing or superficial testing for the random youtube "tech" channel. That's why I'm the only guy you know who's pointed these things out. And if you had done so too then you wouldn't argue with me about what are obvious differences between these TAAUs.
I’ve started playing CP so would really appreciate any settings guides/summaries you recommend or can share because I’ve got mine running pretty much how I want it with path tracing/RT enabled but more by trial and error/blind luck than anything else. :)

I accidentally put FSR on yesterday while experimenting and was confused by the glitching with wire mesh fences and some reflective floors. This stuff is all new to me so could just be my incompetence.
 
I’ve started playing CP so would really appreciate any settings guides/summaries you recommend or can share because I’ve got mine running pretty much how I want it with path tracing/RT enabled but more by trial and error/blind luck than anything else. :)

I accidentally put FSR on yesterday while experimenting and was confused by the glitching with wire mesh fences and some reflective floors. This stuff is all new to me so could just be my incompetence.
What's your system and resolution/hz? Tbh if you're doing PT then that must mean you're on Nvidia, so nothing beats DLSS. With PT (or RT) there's also not much you can do to really "optimise" it further, outside of mods, mostly just setting volumetric clouds & fog to medium, or turning down Crowd Density if your CPU is holding you back. In terms of RT it's really down to three choices. If you want it all, if you want Lighting (Medium for AO, or Psycho for GI boost), or if you want Reflections. Sun Shadows & Local Shadows have very little performance cost once RT is already engaged, f.ex. if you want Lighting or Reflections, so I'd always have them on.

With mods, I'd recommend Ultra+ for RT/PT optimisation and HD Reworked Project for textures. Outside of graphics I personally consider mandatory AutoLoot (huge QoL improvement), Missing Persons (should really be a part of the game almost, it's that good and perfect for if you like losing yourself into the game further lore-wise), and Better Vehicle First Person (it makes actually driving in FP a joy now).
 
What's your system and resolution/hz? Tbh if you're doing PT then that must mean you're on Nvidia, so nothing beats DLSS. With PT (or RT) there's also not much you can do to really "optimise" it further, outside of mods, mostly just setting volumetric clouds & fog to medium, or turning down Crowd Density if your CPU is holding you back. In terms of RT it's really down to three choices. If you want it all, if you want Lighting (Medium for AO, or Psycho for GI boost), or if you want Reflections. Sun Shadows & Local Shadows have very little performance cost once RT is already engaged, f.ex. if you want Lighting or Reflections, so I'd always have them on.

With mods, I'd recommend Ultra+ for RT/PT optimisation and HD Reworked Project for textures. Outside of graphics I personally consider mandatory AutoLoot (huge QoL improvement), Missing Persons (should really be a part of the game almost, it's that good and perfect for if you like losing yourself into the game further lore-wise), and Better Vehicle First Person (it makes actually driving in FP a joy now).
Don't laugh - I'm still dithering over a new monitor at the moment with a view to going 1440p, possibly UW, but for the moment it's just 1920x1200/60Hz. It's obvious that my CPU is bottlenecking as the GPU is only showing about 40% usage at times. I'd like to play with DLDSR for gaming but if you don't have a 16:9 monitor it seems you're stuffed (DLDSR and DSC are not helping me right now given current manufacturer design choices).

I've ended up using DLSS quality with RT and Path tracing and to be honest, whatever it's doing, it looks so good that I forgot I was playing on a low-res monitor. :)

Thanks for those suggestions though, really appreciated and I'll have a play :)
 
Last edited:
@mrk You would have done better to read again for comprehension. I explicitly talked about issues in motion. Who even tests static scenes with TAAUs? Completely pointless.

How FSR is superior to XeSS to how it handles vegetation [for this one wait for 4K to process, can't see it otherwise - or better yet, go test it yourself and verify]:

Or how some issues that exist even in TAA are then exaggerated with XeSS but get suppressed better with FSR (and this is corollary to the visual artifacts that then pop up in certain quest scenes with only XeSS as a result; repeatable in the Q for Meeting Oda, or 6th Street with Panam etc.):

I wish I had a rain scene too, because that's the worst of the bunch, and that's where XeSS is just dog doodoo. Alas, you'll have to make do with these.
{edit- Similar issue but with reflection: https://youtu.be/TV-EjAJjPhI}

I can promise you, there is no one that's spent more time testing Cyberpunk than me, and I've actually played through the whole game with these things, all quests & gigs (and every game version since 1.0, so I know how they all act and how they've changed). Actually playing, not photo mode-ing or superficial testing for the random youtube "tech" channel. That's why I'm the only guy you know who's pointed these things out. And if you had done so too then you wouldn't argue with me about what are obvious differences between these TAAUs.
Thanks although I can't see those in path tracing in my game, plus my screenshots are not taken in photo mode, I have a toggle hotkey to vanish the HUD and that's what I used for the samples I posted. But yes in motion I wouldn't have seen anyway as I didn't test for in motion with the other upscalers generally speaking, DLSS FG/RR is my main use case so that's what I've been focusing on and that's where the full quality is.

I too have completed the game twice now with over 400 hours clocked, the first playthrough was 100%ing the entire map for reference, all using DLSS though mind. I only ever flicked over to FSR/XeSS to see what the image quality was like and compare.

I watched the 4K versions of your videos, I don't know if it's the way AMD cards handle upscalers or something, plus your videos are with RT off and I have no real experience of Cyberpunk with RT off so can't say how motion clarity behaves etc. but here is what I see in motion with FSR vs XeSS vs DLSS, again all in path tracing..

You can direct download the full res raw videos here too, no waiting for anything to process, plus the google drive internal player is always rubbish and maxes out at 1080p which looks like 480p, always download the file direct.

These are all 120fps videos recorded at 150Mbps.

Dry:

Wet:

Vegetation temporal stability:

So yes, I still stand by my previous post(s), XeSS is better than FSR whether in motion or just watching foliage moving in the wind whilst you are static.
 
Last edited:
Back
Top Bottom